* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Export-Led Growth or Growth-Driven Exports? The Canadian Case
Survey
Document related concepts
Transcript
Export-Led Growth or Growth-Driven Exports? The Canadian Case Author(s): Irene Henriques and Perry Sadorsky Reviewed work(s): Source: The Canadian Journal of Economics / Revue canadienne d'Economique, Vol. 29, No. 3 (Aug., 1996), pp. 540-555 Published by: Blackwell Publishing on behalf of the Canadian Economics Association Stable URL: http://www.jstor.org/stable/136249 . Accessed: 03/05/2012 22:49 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. Blackwell Publishing and Canadian Economics Association are collaborating with JSTOR to digitize, preserve and extend access to The Canadian Journal of Economics / Revue canadienne d'Economique. http://www.jstor.org Export-ledgrowth or growth-drivenexports? The Canadiancase IRENE HENRIQUES and PERRY SADORSKY York University Abstract. In this paper we investigate the export-led growth hypothesis for Canadaby constructinga vector autoregression(VAR) in orderto test for Granger(1969) causalitybetween the following variables:real Canadianexports, real CanadianGDP, and real Canadianterms of trade. Two principalresults emerge from our research.First, real Canadianexports, real Canadianterms of trade, and real CanadianGDP are co-integrated.This implies that there exists a long-run steady state among these three variables. Second, we find evidence that a one-way Grangercausal relationshipexists in Canadawhereby changes in GDP precede changes in exports. Croissanceeconomiqueenclencheepar les exportationsou exportationsresultantde la croissance economique?Le cas canadien. Dans ce memoire, les auteursexaminentl'hypothese de la croissance economique enclenchee par les exportationspour le Canadaen construisant une autoregressionde vecteurs pour verifier s'il y a causalite a la Grangerentreles variables suivantes:exportationsreelles canadiennes,PIB reel canadienet termes d'echange reels du Canada. Deux r6sultatsprincipaux s'imposent au terme de cette recherche.D'une part, il semble que les exportationsreelles canadiennes,les termes d'echange du Canadaet le PIB canadienr6el sont co-int6grees. Cela implique qu'il existe un cheminementa long terme en r6gime permanententre ces variables. D'autre part, les auteursmontrentqu'il y a causalite unidirectionnelle'ala Grangerau Canadaet que les changementsdans le PIB pr6cedentles changementsdans les exportations. 1. INTRODUCTION Does rapid income growth lead to rapid trade expansion, or is it the other way around? Anecdotal evidence suggests that countries with strong export performance have strong growth performance and vice versa, Bhagwati (1988) suspects that, as with most economic phenomena, there is a reciprocal relationship between the two economic indicators. The degree to which the relationship is 'genuine' for Canada We thank two anonymousreferees for helpful comments. CanadianJournalof Economics Revue canadienned'Economique, XXiX, No. 3 August aouit 1996. Printed in Canada Imprime au Canada 0008-4085 / 96 / 540-55 $1.50 ? CanadianEconomics Association Export-led growth 541 and the extent to which export growth in Canadadrives GDP growth or vice versa are issues we attemptto address in this paper. Although the idea that trade might influence growth is not new, empirical research exploring the idea has only recently appeared. Relevant literaturecan be divided into two broad categories. The first consists of papers that explore the relationship in developing countries(see Michaely 1977; Balassa 1978; Feder 1982; Jung and Marshall 1985; Chow 1987; Hsiao 1987; Kwan and Cotsomitis 1991). The second includes researchon developed or industrializedeconomies (see Kunst and Marin 1989; Marin 1992; Serletis 1992). Our analysis falls into the second category. With regardsto developed or industralizedcountries,Marin(1992) finds that the hypothesisof export-ledgrowthcannotbe rejectedfor the United States, the United Kingdom,Germany,and Japan.Kunstand Marin(1989) find that the growth-driven export hypothesis cannot be rejectedfor Austria. In this paper we investigate the export-led growth hypothesis for Canada by constructing a vector autoregression(VAR) in order to test for Granger (1969) causality between the following variables: real Canadian exports, real Canadian terms of trade, and real CanadianGDP. Two principal results emerge from our research.First, real Canadianexports, real Canadianterms of trade, and real Canadian GDP are co-integrated.This implies that there exists a long-run steady state among these three variables. Second, we find evidence that a one-way Granger causal relationshipexists in Canadawhereby changes in GDP precede changes in exports. This paper is organizedas follows. In the following section we briefly describe and provide examples of the relationshipbetween exports and GDP. In section III the methodology, the data, and the results are presented.In section IV conclusions are presented. 11. THE RELATIONSHIP BETWEEN EXPORTS AND GDP Three possible relationshipsbetween exports and GDP are examined here: exportled growth, growth-drivenexports, and the two-way causal relationship that we term feedback. Each relationshipwill be discussed in turn. 1. Export-ledgrowth Michaely (1977), Feder (1982), and Marin(1992) found that countriesexporting a large share of their output seem to grow faster than others. The growth of exports has a stimulating influence across the economy as a whole in the form of technological spillovers and other externalities.' Models by Grossman and Helpman (1991), Rivera-Batiz and Romer (1991), and Romer (1990) posit that expanded internationaltrade increases the numberof specialized inputs, increasing growth 1 Increasedexports also can arise from reducedprotectionism.For an excellent discussion regarding protectionismsee Bhagwati (1988). 542 Irene Henriquesand PerrySadorsky rates as economies become open to internationaltrade.2Buffie (1992) considers how export shocks can produce export-led growth. 2. Growth-drivenexports In contrastto the export-led growth hypothesis, scholars such as Bhagwati (1988) have noted that an increase in GDP generally leads to a correspondingexpansion of trade, unless the patternof growth-inducedsupply and correspondingdemand creates an anti-tradebias.3Neoclassical tradetheorytypically stresses the causalitythat runs from home-factorendowmentsand productivityto the supply of exports (see, e.g., Findlay 1984). In the case of Austria, Kunst and Marin(1989) find empirical evidence of growth-drivenexports. The productlife cycle hypothesis developed by Vernon (1966) also has attractedconsiderableattentionamong internationaltrade theorists in recent years. Segerstrom,Anant, and Dinopoulos (1990), for example, use the product life cycle hypothesis as a basis for analysing North-Southtrade in which researchand developmentraces between firms determinesthe rate of product innovationin the North. 3. Feedback The most interesting economic scenarios suggest a two-way causal relationship between growth and trade. According to Bhagwati (1988), increased trade (for whateverreason) produces more income (increasedGDP), and more income facilitates more trade - the result being a 'virtuous circle.' This type of feedback has also been noted by Grossmanand Helpman(1991) in their models of North-South trade. The relationshipbetween GDP growth and export growth is extremely complex, among other reasons because price fluctuationsand political interventionboth influence the relationship.To controlfor these factors the termsof tradeare included. Terms of trade are defined here as export unit value divided by import unit value. Terms of trade are included to control for export growth which results from price competitiveness. Price competitivenessin this case reflects fluctuations in the real exchange rate and possible trade policies (in the form of tariff and non-tariffbarriers).Terms of trade is an especially importantvariable for a small open economy such as Canada, which is highly susceptible to changes in world prices. In fact, recent reports have cited the lower Canadiandollar as one reason Canadahas had higher economic growth than other nations, including the United States.4 2 For a good overview, see Pack (1994). Helpman and Krugman(1985), however, make it clear that the effect of trade on technical efficiency is not conclusive in models of imperfect competition and increasing returnsto scale. In such cases the trade effect depends on the type of competition assumed on the domestic market,entry, and exit and on how marketstructureswill change in response to a tradedisturbance.As a result, the effect of trade on technical efficiency is an empirical issue. 3 Since our data describe an industrializedcountry,the patternof growth-inducedsupply and corresponding demand is likely to be characterizedby a pro-tradebias. 4 See the Globe and Mail, 'Canadabucks the trend,' 1 June 1993. Export-led growth 543 111. METHODOLOGY Whetherexports cause GDP gains or losses, whether GDP gains cause exports, or whether a two-way causal relationshipexists between exports and GDP can, in the end, be decided only empirically. Our investigation of the relationships between Canadian exports and various other macroeconomicvariablesproceeds by studying the integrationpropertiesof the data and undertakinga systems co-integratinganalysis and examining Granger causality tests. In order to check the temporal stability of the GDP, exports, and terms of trade relationships, our analysis is performed over two subsamples in addition to the full sample. The first period, 1877-1945, was chosen so as to include the first two world wars and the Great Depression. The second period, 1946-91, is the postwarperiod. 1. The data The data are annual Canadianobservationson real GDP,5 real exports, and real terms of trade defined as export unit value divided by import unit value. Annual data on all variablesare availablefrom 1870 to 1991. Data definitions and sources are listed in the data appendix. Plots of the logarithms of the three time series are shown in figures 1 to 3. Figures 1 and 2 demonstratethatboth the naturallogarithmof real GDP, lrgdp,and naturallogarithmof real exports, lrex, exhibit strong upwardtrends.This provides anecdotalevidence that the two series tend to move together. Figure 3 is a plot of the naturallogarithmof real terms of trade, lrtt. This series is extremely volatile, reflecting possible linkages of real exchange rate fluctuationsas well as possible effects of trade policy. Summary statistics of lrgdp, lrex, and lrtt indicate that these variables have means equal to 6.0796, 4.3539, and -0.1903 withlassociated standarddeviations of 1.3785, 1.4674, and 0.1681, respectively. 2. Integrationproperties of the data In order to investigate the stationaritypropertiesof the data, a univariateanalysis of each of the three time series (real GDP, real exports, and real terms of trade) was carriedout by testing for the presenceof a unit root. AugmentedDickey-Fuller (ADF) t-tests (Dickey and Fuller 1979) and Phillips and Perron(1988) Z(t&x)-tests for the individual time series and their first differences are shown in table 1. The lag length, k, for the ADF tests was selected to ensure that the residualswere white noise. We first consider the levels of the series of the three variables, namely, the naturallogarithmsof (a) real GDP (lrgdp),(b) real exports (rex), and (c) real terms of trade (lrtt). It is obvious from the ADF and Phillips and Perron tests that at conventional levels of significance, none of the variables represents a stationary 5 Real GDP has some advantagesover the use of GNP in that it is more closely related to employment within Canadaand easier to reconcile with measurementsof output by province (Hall, Taylor, and Rudin 1990). 544 Irene Henriquesand PerrySadorsky 9 8.5 8 7.5 7 6.5 6 5.5 5 4.5 4 1870 1878 1886 1894 1902 1910 1918 1926 1934 1942 1950 1958 1966 1974 1982 1990 1874 1882 1890 1898 1906 1914 1922 .1930 1938 1946 1954 1962 1970 1978 1986 1994 Year FIGURE 1 Graph of the series Irgdp 7.5 7 6.5 6 5.5 5 4.5 I4 3.5 3 2.5 1870 1878 1886 1894 1902 1910 1918 1926 1934 1942 1950 1958 1966 1974 1982 1990 1874 1882 1890 1898 1906 1914 1922 1930 1938 1946 1954 1962 1970 1978 1986 1994 Year FIGURE 2 Graph of the series Irex Export-led growth 545 0.1 0 -0.1 -0.2 04~~~~~~~~~~ -0.3 -0.6 -0.7 I IE I I I i , lI l I I I -0.8 E i a 1870 1878 1886 1894 1902 1910 1918 1926 1934 1942 1950 1958 1966 1974 1982 1990 1874 1882 1890 1898 1906 1914 1922 1930 1938 1946 1954 1962 1970 1978 1986 1994 Year FIGURE 3 Graph of the series Irtt process. ADF and Phillips and Perrontests computed using the first difference of lrgdp, Irex, and lrtt indicate that these tests are individuallysignificantat the 1 per cent level of significance.As differencingonce produces stationarity,we conclude that each of the series lrgdp, lrex, and lrtt is integratedof order 1 (1(1)). 3. Systems co-integratinganalysis We begin the analysis with a congruentstatistical system of unrestrictedreduced forms representedby equation(1). p Yt - - + 2_;lIyt-, + ct, c-t- IN(O, Q), t = 1 ... .,T, 1 whereyt is an (n x 1) vector of 1(1) and/or1(0) variablesand fz is an n x 1 vector of constants. Letting Ayt =yt --yt-., a convenient reparameterizationof (1) is given by (2). p-l Ay IL+ E H,Ayt, + flyt-p + ct, (2) T-i where both LI-Iand l = P1 FIr-I are of dimensionn x n. This is the JiT1=X vector autoregression(VAR) approachthat Johansen (1988, 1991) and Johansen and Juselius (1990) used to investigate the co-intergrationpropertiesof a system. 546 Irene Henriquesand PerrySadorsky TABLE 1 Unit root tests Augmented Dickey-Fuller(Dickey and Fuller 1979) test regression k Ayt = t + ayt-I + E CkAyt-k + Ut, t =1877-1991 k=l Variable ADF k lrgdp lrex Irtt -0.540 -0.311 -0.276 1 3 5 Alrgdp Alrex Alrtt -5.842*** -6.166*** -6.462*** 4 4 5 Phillips and Perron(1988) test regression Yt = A + ayt-, Variable + ut, t = 1877-1991 Z(t&) I Irgdp Irex lrtt 0.383 0.043 -2.279 9 9 6 Alrgdp Airex Alrtt -8.570*** -9.034*** -11.239*** 10 9 6 NOTES ***, **, and * denote that a test statistic is significant at the 1 per cent, 5 per cent, and 10 per cent levels of significance, respectively. The 1 per cent, 5 per cent, and 10 per cent critical values, from Fuller (1976, 373) are -3.51, -2.89, and -2.58, respectively. 1 is a truncatedlag parameter used in the non-parametriccorrection for serial correlationand is set as the highest significant lag order from either the autocorrelationfunction or the partialautocorrelationfunction of the first differencedseries. The lag length, p, is chosen to ensure that the errorsare iid. Since Et is stationary, the rank, r, of the 'long-run' matrix fl determineshow many linear combinations of y, are stationary.If r = n, all yt are stationary,while if r = 0 so that lI = O, Ay, is stationaryand all linear combinationsof yt - (1). For 0 < r < n there exist r co-integratingvectors, meaning r stationarylinear combinationsof Yt. In that case, III can be factored as a(x', where both a and 3 are n x r matrices. The co-integratingvectors of 3 are the errorcorrectionmechanismsin the system, while a contains the adjustmentparameters.This result is known as Granger's RepresentationTheorem and can be found in Engle and Granger(1987). Johansen and Juselius (1990) provide a full maximum likelihood procedurefor estimation and testing within this framework. Johansen's procedurewas originally derived under the assumptionsof normal errorsand an optimal lag length,p, for the VAR shown in equation(2). In practice Export-led growth 547 the question arises as to how sensitive the Johansen procedure is to the choice of lag length and non-normality.Using Monte Carlo techniques, Gonzalo (1994) compares cointegratingparameterestimates from OLS, NLS, and MLE in an error correction model and finds that all parameterestimates from each of these three estimation methods are super consistent. He also finds that maximum likelihood estimation in a fully specified errorcorrectionmodel has betterpropertiesthan the other methods even when the errors are non-Gaussianor when the dynamics are unknown. The co-integratingrank, r, can be formally tested with two statistics. The first is the maximum eigenvalue test. Denoting the estimated eigenvalues as Ai, i = 1, 2, ... , n, the maximum eigenvalue test is given by Amax= -T ln (1 - Ar+l), (3) where the appropriatenull is r = g co-integratingvectors against the alternative that r < g + 1. The second statistic is the trace test and is computed as n Trace =-T E In (1- Ai) (4) i=r+1 where the null being tested is r = g against the more general alternativer _ n. The distributionof these tests is a mixture of functionals of Brownian motions that are calculated via numericalsimulationby Johansen and Juselius (1990) and Osterwald-Lenum(1992). Cheungand Lai (1993) use Monte Carlomethodsto investigatethe small sample properties of Johansen's Amax and trace test statistics. In general, they find that both the Amax and trace test statistics are sensitive to underparameterizationof the lag length although they are not so to overparameterization. They suggest that information criteria such as Akaike's (1974) AIC or Schwarz's (1978) BIC can be useful in determiningthe correct lag length. They also find that the trace test statistic is more robust to both skewness and excess kurtosis than the Amax test. Variousorderselection criteriawere appliedto unrestrictedVAR models in order to determinethe appropriateorder of the VAR. The order selection criterion does not require stationarityof the system (see Liitkepohl 1991). The Akaike (1974) AIC, Schwarz (1978) BIC, and Hannanand Quinn (1979) HQ criteriawere used to determinethe order of the VAR. The AIC, BIC, and HQ criteriachose p = 4, p = 2, andp = 3, respectively.A VAR with p = 2 exhibited a high degree of serial correlation in the residuals an(dwas therefore excluded as a possible candidate. VARs with both p = 3 and p = 4 demonstratedwhite-noise residuals. In the interestof parsimony,a VAR with lag lengthp = 3 was chosen. To investigate the sensitivity of our results with respect to choice of lag length, we also reportresults from models estimatedwith p ==4. The results of Johansen's tests for co-integrationfor the three sample periods with VAR lag lengthsp = 3 andp = 4 are presentedin table 2. When the lag length 548 Irene Henriquesand PerrySadorsky TABLE 2 Tests for co-integrationusing the Johansenprocedure p-1 fl,Ay,_ + nytp Ayt = Lj+ -n = a/3' + E,, r1 1877-1991,p = 3 eigenvalues The test statistics Hypothesis 0.1714 r= 0 r_ 1 r_ 2 Trace test A max test 36.092*** 21.627** 14.465* 14.372** 0.093 0.093 0.1299 0.0000 0.1175 0.0008 Co-integratingequation lrgdp = 0.761-Irex + 2.637-1rtt - 3.227 1877-1945, p = 3 eigenvalues The test statistics Hypothesis 0.2568 r= 0 Trace test A max test 30.089** 20.484* r< 1 r< 2 9.609 9.606 0.002 0.002 0.1977 0.0599 Co-integratingequation lrgdp = 0.802-Irex + 1.433-1rtt - 2.795 1946-1991,p = 3 eigenvalues The test statistics 0.5057 Hypothesis r=0 r< 1 r< 2 Trace test A max test 45.389*** 32.413*** 12.976 10.134 2.842 2.842 Co-integratingequation lrgdp = 0.736-Irex - 1.454-Irtt - 3.149 1877-1991, p = 4 eigenvalues The test statistics Hypothesis 0.1603 r=0 r< 1 r_ 2 Trace test A max test 33.828** 20.096* 13.731 13.718* 0.013 0.013 Co-integratingequation lrgdp = 0.678-Irex + 3.737-1rtt - 3.784 0.1125 0.0001 Export-led growth 549 TABLE 2 (Concluded) 1877-1945,p = 4 0.1313 0.2069 eigenvalues 0.0020 The test statistics Hypothesis r= 0 Trace test A max test 25.854 16.003 r 1 r 2 9.851 9.712 0.139 0.139 0.1909 0.0503 Co-integratingequation lrgdp = 0.779-lrex + 1.540-lrtt- 2.899 1946-1991, p = 4 eigenvalues 0.5699 The test statistics Hypothesis r= 0 r 1 Trace test A max test 50.939*** 38.821*** 12.119 9.746 r 2 2.372 2.372 Co-integratingequation lrgdp = 0.788-Irex - 1.779-Irtt- 2.834 NOTES *, * and * denote that a test statistic is significant at the 1 per cent, 5 per cent, and 10 per cent levels of significance, respectively. Critical values are taken from Osterwald-Lenum(1992, table 1). is set at p 3, both the trace and A max test statistics indicatethat a co-integration rank of one is present in each of the three sample periods. Consequently,lrgdp, lrex, and lrtt are co-integrated.When the lag length is set at p = 4, co-integration is establishedfor the periods 1877-1991 and 1946-91 but not for the period 18771945. Our results indicate that with longer lag lengths in the VAR it is easier to establish co-integrationover the periods 1877-1991 and 1946-91 than over the period 1877-1945. 4. Granger causality tests It is interesting to analyse the Grangercausality structureof the three variables considered.According to a theoremdeveloped by Sims, Stock, and Watson(1990), such an analysis is very simple if the variables are cointegrated,as in our case, because the standardF-statisticsfor the hypothesis ij,= O, Tr= 1, 2,. .., p in the level VAR have asymptoticF-distributionsin spite of the fact thatthey are obtained within the frameworkof an I(1) system. Probabilityvalues for various F-statistics are presentedin table 3. The results reported in table 3 indicate that for all cases, changes in lrgdp Grangercause changes in lrex. For the three sample periods consideredwith p = 3, 550 Irene Henriquesand PerrySadorsky TABLE 3 Grangercausality tests p Yt = A + E nryt-i- + Et F-statistics (Ho: nij, = 0, T = 17 2, ... ,p) y'= [lrgdp, Irex, lrtt] 1877-1991, p = 3 CausationFrom: Irgdp Irex lrtt To: lrgdp Irex lrtt 0.3179 0.6555 0.0001*** 0.3732 0.1130 0.0054*** 1877-1945, p - 3 CausationFrom: lrgdp Irex lrtt 1946-1991,p To: Irgdp Irex lrtt 0.5376 0.6809 0.0038*** 0.6767 0.0861* 0.0044*** = 3 CausationFrom: lrgdp Irex lrtt To: lrgdp Irex lrtt 0.5796 0.0158** 0.0016*** 0.1139 0.6399 0.5480 - 1877-1991, p = 4 CausationFrom: Irgdp Irex Irtt To: lrgdp Irex Irtt 0.5752 0.7191 0.0022*** 0.6416 0.1927 0.0124** Export-led growth 551 TABLE 3 (Concluded) 1877-1945, p = 4 CausationFrom: Irgdp Irex Irtt To: Irgdp Irex Irtt 0.6873 0.6582 0.0528* 0.9026 0.1895 0.0141** 1946-1991,p = 4 CausationFrom: Irgdp Irex Irtt To: Irgdp Irex Irtt 0.8095 0.0419** 0.0100*** - 0.4869 0.3994 0.2556 - NOTES Table 3 shows probabilityvalues. * * and * denote that a test statistic is significant at the 1 per cent, 5 per cent, and 10 per cent levels of significance, respectively. the hypothesis that changes in lrgdp do not Grangercause changes in Irex can be rejected at the 1 per cent level of significance. For p = 4 the hypothesis that changes in lrgdp do not Grangercause changes in Irex can be rejected at the 1 per cent level of significance for the sample periods 1877-1991 and 1946-91, while the hypothesiscan be rejectedat the 10 per cent level of significance for the period 1877-1945. This provides strong evidence in favour of the growth-drivenexport hypothesis. Also of interestin table 3 is the fact that for the samples 1877-1991 and 18771945, lrex exhibits Grangercausalityon lrtt.This suggests that historically Canada may not be the small open economy that it is often thought of. One possible explanation for this result is that Canada is natural resource abundantand that while naturalresourcescurrentlyaccountfor about50 per cent of exports(Anderson 1985, 13) it is also the case that historically naturalresources have representeda large share of exports. Moreover,fifty to a hundredyears ago, Canadawas a larger player in world resource markets, suggesting that Canada's role in international marketsduringthe period 1877-1945 may have been greaterthan its role today. Since Serletis (1992) is the only paper,to our knowledge, thathas used Canadian data to explore the relationshipbetween exports and GNP, a few words regarding his approachand how it differs from ours are necessary. Serletis (1992) uses annual Canadiandata (exports, imports, and GNP) from 1870 to 1985 to explore the relationshipbetween export growth and economic growth. With single-equation techniques, he suggested that growth in GNP and 552 Irene Henriquesand Perry Sadorsky export growth are independent. Serletis's findings support an export-led growth strategy in that expansion in exports is described as promotinggrowth in national income. Our approachdiffers from Serletis (1992) in two fundamentalways. First, we include a terms-of-tradevariable to account for price competitiveness.Second, we employ a multivariateestimation methodology which takes explicit account of possible feedback and simultaneityeffects as well as the long-runpropertiesof the data.6 In summary,our results tend to favour the one-way Grangercausal relationship that suggests that the growth rate of GDP influences export growth. IV. CONCLUSIONS Our objective was to test the causal relations implied by three competing, although not mutually exclusive, hypotheses: (1) the export-led growth hypotheses; (2) the growth-drivenexports hypothesis; and (3) the two-way causal hypothesis, which is a combinationof (1) and (2). Before we tested the various relationshipsit was necessaryto determinewhether lrgdp,lrex, and lrttwere co-integrated.We found thatthey were, implying thatthere exists a long-run(or steady-state)relationshipbetween them. Empirical evidence from a VAR suggests that the growth-drivenexports hypothesis cannot be rejected.There is no evidence supportingthe export-ledgrowth hypothesis. It is importantto realize that our study of Grangercausal orderings of export and GDP does not identify causal directions but instead asks the question whether movementsin exports tend to precede or to follow movementsin GDP. Our empirical results for Canada,like those of Kunst and Marin(1989) for Austria, suggest that changes in growth precede changes in exports. This is in accord with the development of a small open economy, since a small economy developing efficiently in line with its comparative advantagewill specialize and hence turn to foreign marketsfor exports of goods that use its most abundantfactor of productionmost intensively. DATA APPENDIX 1. The termsof tradeis the export price as a percentageof importprice. Data from 1870 through 1991 were collected. Source: 1870-1975: Leacy, F.H. (1983) Historical Statistics of Canada, second edition, Catalogue 11-516, table G386-388 (Ottawa:Statistics Canada) The table in question had various base years, but given that each subseries overlapped,the SPLICEcommandin SHAZAMwas employed.The years 18701975 were convertedin terms of a 1971 base (i.e., 1971 = 100). 6 Serletis (1992, 134) acknowledges a possible simultaneityproblem.This could give rise to misleading results. Export-ledgrowth 553 Source: 1976-91: Statistics Canada,CANSIM (in conformancewith the Leacy data, the Laspeyresimport and export price indices were retained).The specific sources are as follows: import price index 1971 = 100: Matrix 3681 D395700 import price index 1981 = 100: Matrix 3635 D448729 import price index 1986 = 100: Matrix 3622 D750816 export price index 1971 100: Matrix 3684 D399017 export price index 1981 = 100: Matrix 3638 D449761 export price index 1986 = 100: Matrix 3625 D751848 Once again, the SPLICEcommand in SHAZAM was employed to convert the entire terms of trade series (1870 through 1991) into a 1981 base year. 2. GDP in millions of currentCanadiandollars 1870-1985 Source: Urquhart,M.C. (1988) Canadianeconomic growth, 1870-1980.' Discussion PaperNo. 734, Departmentof Economics, Queen's University For 1986 to 1991 we made use of Statistics Canada, CANSIM matrix series # 6917 D 11908. 3. Exports in millions of currentCanadiandollars. Source: 1870-1975: Leacy, F.H. (1983), Historical Statistics of Canada, second edition, Catalogue 11-516 (Ottawa:Statistics Canada) Source: 1976-91: Exports -- Statistics Canada, CANSIM matrix series # 2333 D 70009 4. All values expressed in currentCanadiandollars were expressed in real terms using the implicit Price Index (1981 - 100). Source: 1870-1985: Urquhart,M.C. (1988) 'Canadianeconomic growth, 18701980.' Discussion PaperNo. 734, Departmentof Economics, Queen's University Source: 1986-91: Statistics Canada, CANSIM matrix series # 6120A 48662. Since after 1986 the base year was set to 1986 - 100, data for 1987-1991 were adjustedto 1981 = 100 using Statistics Canada,CANSIM matrixseries # 6841 D 14477. REFERENCES Akaike, H. (1974) 'A new look at the statisticalmodel identification.'IEEE Transactions on AutomaticControl, AC-19, 716-23 Anderson, F.J. (1985) Natural Resources in Canada, Economic Theory and Policy (Toronto:Methuen) Balassa, B. (1978) 'Exports and economic growth:furtherevidence.' Journal of DevelopmentEconomics 2, 181-9 Bhagwati,J. (1988) Protectionism(Cambridge,MA: MIT Press) 554 Irene Henriques and Perry Sadorsky Buffie, E.F. (1992) 'On the condition for export-led growth.' This Journal 25, 211-25 Cheung, Y.W., and K.S. Lai (1993) 'Finite sample sizes of Johansen's likelihood ratio tests for cointegration.'OxfordBulletin on Economics and Statistics 55, 313-28 Chow, P.C.Y. (1987) 'Causality between export growth and industrialperformance:evidence from the NIC's.' Journal of DevelopmentEconomics 26, 55-63 Dickey, D.A., and W.A. Fuller (1979) 'Distributionof the estimatorsfor autoregressive time series with a unit root.' Journal of the American StatisticalAssociation 74, 42731 Engle, R.F., and C.W.J. Granger(1987) 'Cointegrationand errorcorrection:representation and testing.' Econometrica 55, 251-76 Feder, G. (1982) 'On exports and economic growth.'Journal of DevelopmentEconomics 12, 59-73 Findlay,R. (1984) 'Growth and developmentin trade models.' In Handbook of International Economics, vol. 1, ed. R. Jones and P. Kenen (Amsterdam:North-Holland) Fuller,W.A. (1976) Introductionto Statistical Time Series (New York: Wiley) Gonzalo, J. (1994) 'Five alternativemethods of estimatinglong-runequilibriumrelationships.' Journal of Econometrics 60, 203-34 Granger,C.W.J. (1969) 'Investigatingcausal relationshipsby econometricmodels and cross-spectralmodels.' Econometrica 37, 424-38 Grossman,G.M., and E. Helpman (1991) Innovationand Growthin the Global Economy (Cambridge,MA: MIT Press) Hall, R.E., J.B. Taylor, and J.R. Rudin (1990) Macroeconomics:The Canadian Economy, first edition (New York: Norton) Hannan,E.J., and B.G. Quinn (1979) 'The determinationof the order of an autoregression.' Journal of the Royal Statistical Society B, 41, 190-5. Helpman,E., and P. Krugman(1985) MarketStructureand Foreign Trade(Cambridge, MA: MIT Press) Hsiao, M.C.W. (1987) 'Tests of causality and exogeneity between exports and economic growth:the case of the Asian NIC's.' Journal of Economic Development 12, 14359 Johansen,S. (1988) 'Statistical analysis of cointegrationvectors.' Journal of Economic Dynamics and Control 12, 231-54 (1991) 'Estimationand hypothesis testing of cointegratingvectors in Gaussianvector autoregressivemodels.' Econometrica 59, 1551-80 Johansen,S., and K. Juselius (1990) 'Maximumlikelihood estimationand inference on cointegration- with applicationto the demand for money.' OxfordBulletin of Economics and Statistics 52, 169-210 Jung, W.S., and P.J. Marshall(1985) 'Exports, growth and causality in developing countries.' Journal of DevelopmentEconomics 18, 1-12 Kunst, R., and D. Marin(1989) 'On exports and productivity:a causal analysis.' Review of Economics and Statistics 71, 699-703 Kwan, A.C.C., and J. Cotsomitis (1991) 'Economic growth and the expanding export sector: China 1952-1985.' InternationalEconomic Journal 5, 105-17 Leacy, F.H. (1983) Historial Statistics of Canada, second edition, Catalogue 11-516 (Ottawa:Statistics Canada) Lutkepohl,H. (1991) Introductionto Multiple Time Series Analysis (Berlin: SpringerVerlag) Marin,D. (1992) 'Is the export-led growth hypothesis valid for industralizedcountries?' Review of Economics and Statistics 74, 678-88 Michaely, M. (1977) 'Exports and growth: an empirical investigation.'Journal of DevelopmentEconomics 40, 49-53 Osterwald-Lenum,M. (1992) 'A note with quantilesof the asymptoticdistibutionof the Export-led growth 555 maximumlikelihood cointegrationrank test statistics: four cases.' OxfordBulletin of Economics and Statistics 54, 461-72. Pack, H. (1994) 'Endogenousgrowth theory: intellectual appeal and empirical shortcomings.' Journal of Economic Perspective 8, 55-72 Phillips, P.C.B., and P. Perron(1988) 'Testing for a unit root in time series regression.' Biometrika75, 335-46 Rivera-Batiz,L., and P.M. Romer (1991) 'Economic integrationand endogenous growth.' QuarterlyJournal of Economics 106, 531-56 Romer, P.M. (1990) 'Endogenoustechnological change.' Journal of Political Economy 98, s71-s102 Schwarz, G. (1978) 'Estimatingthe dimension of a model.' Annals of Statistics 6, 461-4 Segerstrom,P.S., T.C.A. Anant, and E. Dinopoulos (1990) 'A Schumpeterianmodel of the product life cycle.' AmericanEconomic Review 80, 1077-91 Serletis, A. (1992) 'Export growth and Canadianeconomic development.'Journal of DevelopmentEconomics 38, 133-45 Sims, C.A., J.H. Stock, and M.W. Watson (1990) 'Inference in linear time series models with some unit roots.' Econometrica58, 113-44 Urquhart,M.C. (1988) 'Canadianeconomic growth, 1870-1980.' Discussion Paper No. 734, Departmentof Economics, Queen's University Vernon, R. (1966) 'Internationalinvestmentand internationaltrade in the productcycle.' QuarterlyJournal of Economics 80, 190-207